Efficient Neural Networks
Ishan discusses the evolution of neural network architectures, highlighting the effectiveness of RegNets in balancing computational efficiency and memory usage. He emphasizes the importance of considering both Flops and memory when designing networks, revealing how innovations like squeeze excitation blocks enhance performance. The conversation also touches on the transition from convolutional networks to Transformers for vision tasks, showcasing the versatility needed in modern AI applications.In this clip
From this podcast
Lex Fridman Podcast
Ishan Misra: Self-Supervised Deep Learning in Computer Vision | Lex Fridman Podcast #206
Related Questions